What Data for This Category Consists Of
This category of data mainly comes from local public data open platforms, integrated big data platforms, departmental business databases, and electronic certificate repositories. The update frequency is once per month. Routinely processed documents include public data resource catalogs, data openness lists, data item descriptions, desensitization rules, interface documents, and update frequency descriptions. The content covers structured fields such as data item names, open scope, desensitization requirements, calling interface paths, and update cycles. Some documents also include interface calling examples and permission authorization boundary descriptions.
Constraints on Knowledge Base Retrieval and Recall
The monthly update cycle of the data requires the knowledge base to be configured with regular synchronization tasks. This prevents retrieval results from being inconsistent with the latest open data. Structured fields and interface documents require retrieval to support precise matching of fields such as data item names and interface paths, while also associating desensitization rules and open conditions. Data from multiple sources may have format differences. Field hierarchy must be retained during knowledge base chunking to ensure retrieval semantic accuracy. Data leakage risks require retrieval results to automatically filter undesensitized raw data, and only return content that complies with open permissions.
Configuration Settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.65–0.75 | Public data catalog fields mostly require precise matching. A threshold that is too low will introduce irrelevant data items, while a threshold that is too high may miss eligible open data |
limit | Top 8–12 results | Declaration materials and file review need to cover multi-dimensional data items. Returning too many results in a single retrieval will exceed the context window, while returning too few will fail to meet complete review requirements |
searchMode | vector | Prioritize semantic retrieval to match data item descriptions and review requirements, combined with structured fields to supplement precise matching |
usingReRank | Enabled | Public data documents contain a large number of similar fields. Reranking can filter low-correlation results and improve review accuracy |
datasetSearchUsingExtensionQuery | Enabled | Automatically expand declaration requirements during review into data item keywords to cover a wider range of open data catalog content |
collectionIds | Bind the public data catalog collection corresponding to the department | Limit the retrieval scope to the open data resources of the corresponding department, avoid confusion of cross-department data, and ensure review compliance |
Three Common Misconfigurations
- Semantic retrieval scores display values over 4000, falling outside the standard 0–1 range: This happens when the normalization logic for similarity calculation is not properly configured. Raw vector distances are returned directly without conversion to standardized scores.
- Retrieval results do not return associated information for corresponding data files: This occurs when the knowledge base file metadata binding configuration is not enabled, or the
collectionIdsparameter is not correctly specified to limit the retrieval scope. This prevents association of document sources. - Retrieval results do not update after knowledge base synchronization: This happens when the scheduled synchronization task is not configured for the monthly update frequency, or the synchronization task trigger parameters are set incorrectly. This causes data to not refresh in a timely manner.
How to Verify Correct Configuration
- Initiate a test query that includes a specific data item name, and check whether the
similarityscore of the returned results falls within the preset range. - Review the retrieval result list to confirm the number of returned results matches the set range, and that no irrelevant data from non-specified departments is included.
- Submit a query that includes an interface path to verify that the reranking function filters low-correlation results, and only retains documents related to the interface.
- Manually trigger a knowledge base synchronization, wait 10 minutes, then run the query again. Confirm that the retrieval results include the latest updated data item content.
The values provided here are common starting points and should be measured against the reader's own samples.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-28.